⚡

Polling Error — AI Polling Intelligence System

Program: Polling Error • Course: LIAI 1001 • Georgian College

✨ Prompter Studio 📥 Download StatBot PDF
9 Agents Active • Ready
Student: Graham Turnbull (200679442)
Program: "Polling Error" • 9 Specialized Agents + HITL Gatekeeper • Causal Audit Engine

Bridging the Gap Between
Polling Data & Electoral Reality

Powered by PollBot (Anthropic Claude qualitative reasoning + SerpApi), StatBot (taxonomic categorization + ReportLab PDF), and an Orchestrator-Worker Pipeline with mandatory Human-in-the-Loop oversight.

📥 Ingest Polling Data to Analyze

Submit custom survey parameters, load a verified benchmark preset, paste raw JSON, or upload a dataset file to trigger the multi-agent system.

Multi-Agent Pipeline Ready
Retrospective Election Outcome (Optional — for Divergence Audit)

Select a verified benchmark poll from the catalog below to immediately load and analyze its statistical properties, causal mechanisms, and taxonomy:

📄 Ingest & Audit PDF Polling Report

The "Polling Error" PDF engine extracts document text page-by-page, verifies the Domain Relevance Guardrail, isolates methodology metrics, and builds an intelligence dossier.

📑

Drop PDF Poll Report Here or Click to Browse

Upload survey releases, poll memos, or academic audit briefs (.pdf format).

⚡ Or Select a System Publication PDF Dossier to Ingest:

Paste a polling record in JSON format to ingest and process:

📁

Drag & Drop Poll Dataset Here

Supports .pdf, .json, .csv, or .txt poll files.

📡

System Ready — No Poll Loaded Yet

The "Polling Error" multi-agent analytical system does not display unprompted statistics on open. Use the Ingestion Portal above to submit a new survey, select a verified benchmark, or upload a dataset to begin the qualitative and taxonomic analysis.

Quick Benchmark Launch:
✨ Prompter on "Polling Error" • Multi-Agent Guiding Interface

The Prompter on "Polling Error"

Configure, validate, and inject structured guiding instructions for PollBot, StatBot, and the Sequential Pipeline. Includes real-time guardrail pre-screening, category partitioning rules, and a live broadcast teleprompter.

Select a Prompter Recipe or Craft Custom Directive:
🛡️ Guardrail Pre-Flight: ✅ Objective & Evidence-Based ✅ Domain Verified (Polling / Surveys) 🔒 Reaffirmation Gate Active
ON AIR • LIVE TELEPROMPTER VIEW
Scroll Speed: 1x
"Focus on PollBot results and systematically categorize into the four requested groups..."
⚖️ Orchestrator-Worker Pipeline • Bayesian Consensus & Human-in-the-Loop (HITL) Gate

Multi-Poll Consensus Aggregator & Human Oversight Checkpoint

Execute Agents 1–5 (Orchestrator → Normalizer → Modeler → Explainer → HumanAuditor) across multiple concurrent surveys to compute Bayesian consensus and review flagged outliers (>3.5% divergence).

Poll ID Pollster Organization Sample Size & Frame Collection Mode Candidate A Candidate B Normalized Weight Outlier Status
POLL-001 National Metric Research N = 1,200 • Likely Voters (LV) Live Phone + SMS 49.2% 47.8% 1.15x (High Rigor)
POLL-002 Civic Pulse Online N = 850 • Registered Voters (RV) Online Opt-in Panel 45.1% 52.4% 0.72x (Discounted)
POLL-003 Public Opinion Daily N = 1,100 • Likely Voters (LV) Mixed Mode (IVR + Online) 48.8% 48.1% 0.96x (Standard)
Modeled Consensus (Cand A)
48.1%
Modeled Consensus (Cand B)
49.0%
Aggregate Margin Spread
0.9 pp
Outliers Flagged (>3.5%)
1 (POLL-002)
Pipeline State
HUMAN_CHECKPOINT
🛡️

Agent 5: Human Oversight Checkpoint (HITL Gatekeeper)

Triggered automatically: Civic Pulse Online (POLL-002) diverged beyond the 3.5% outlier threshold due to Registered Voter (RV) screen and Online Opt-in Panel mode skew.

🤖 Multi-Agent System Architecture • Complete 9-Agent Roster (AGENTS.md)

The 9 Autonomous Agents of "Polling Error"

Inspect every specialized agent in the system—from raw ingestion and Bayesian modeling to Claude qualitative reasoning, ReportLab taxonomy publishing, and human oversight.

⚡

Try Out "Polling Error" Live Multi-Agent Console

Interact with PollBot, StatBot, the Sequential Pipeline, and the Collaborative Loop with live guardrail enforcement

🔒 Handoff Reaffirmation Gate:
Engine: Deterministic Simulation / Grounded Knowledge
Console Telemetry & Agent Output:
Ready. Select a test preset above or type your prompt to try out Polling Error!

📬 Contact & Polling Intelligence Inquiry Desk

Submit methodology audit requests, dataset submissions, governance inquiries, or academic evaluation feedback routed directly to the "Polling Error" multi-agent triage desk.

🔒 Zero-PII Retention Enforced
Project & Academic Directory
System Lead & Student Author
Graham Turnbull
Student ID: 200679442
Academic Course & Institution
LIAI 1001 – Strategy and Change
AI Leadership & Management Program • Georgian College
Instructor: Prof. Evandro Barros
Automated Multi-Agent Routing
• Methodology & Mode Bias: Routed to PollBot (Agent 6)
• Taxonomy & PDF Dossiers: Routed to StatBot (Agent 7)
• Outlier >3.5% Governance: Routed to HumanAuditor (Agent 5)
🗄️ SQLite3 & Cloudflare D1 Database Persistence

Submitted contact inquiries are automatically validated against Predetermined Conclusion guardrails, signed with a SHA-256 receipt, and saved into the relational database (data/contact_inquiries.db).

🗄️ Saved Contact Inquiries Database Ledger (contact_inquiries Table)

Live SQLite3 / Cloudflare Edge SQL records stored from submitted contact forms (SQLite3: data/contact_inquiries.db).

1 Record Saved
Receipt ID Timestamp (UTC) Sender & Email Organization Assigned Agent Desk Subject & Inquiry Message
Loading saved contact inquiries from database...